Hindawi

Journal of Applied Mathematics

Volume 2020, Article ID 5109841, 13 pages
https://doi.org/10.1155/2020/5109841

Research Article
Mathematical Modeling, Analysis, and Optimal Control of
Corruption Dynamics

Haileyesus Tessema Alemneh

Department of Mathematics, College of Natural and Computational Sciences, University of Gondar, Gondar, Ethiopia

Correspondence should be addressed to Haileyesus Tessema Alemneh; haila.tessema@gmail.com

Received 20 March 2020; Revised 16 May 2020; Accepted 8 June 2020; Published 1 August 2020

Academic Editor: Fernando Simões

Copyright © 2020 Haileyesus Tessema Alemneh. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work
is properly cited.

In this paper, a nonlinear deterministic model for the dynamics of corruption is proposed and analysed qualitatively using the
stability theory of differential equations. The basic reproduction number with respect to the corruption-free equilibrium is
obtained using next-generation matrix method. The conditions for local and global asymptotic stability of corruption-free and
endemic equilibria are established. From the analysis using center manifold theory, the model exhibits forward bifurcation.
Then, the model was extended by reformulating it as an optimal control problem, with the use of two time-dependent controls
to assess the impact of corruption on human population, namely, campaigning about corruption through media and
advertisement and exposing corrupted individuals to jail and giving punishment. By using Pontryagin’s maximum principle,
necessary conditions for the optimal control of the transmission of corruption were derived. From the numerical simulation, it
was found that the integrated control strategy must be taken to fight against corruption.

1. Introduction tion.The study by Abdulrahman [7] proposed and analysed a

Corruption is an illegal activity carried out for private gain
and benefit, by misuse of authority or power by public
(government) or private (company) officeholders [1]. Differ-
ent types of corruption are documented in the literature and
include public corruption [2], private corruption [3], perva-
sive corruption, and arbitrary corruption [4]. Corruption
can originate from either the demand side or the supply side
[5]. In general, corruption faces a major threat to the rule of
law, democracy and human rights, fairness, and social justice,
hinders economic development, and brings market econo-
mies at risk for their proper and fair functioning [2, 6]. It is
a serious problem in all countries in the world mainly for
developing countries [6]. Even though most countries have
anticorruption policies or strategies that are being made to
control corruption, it remains an epidemic in the society.
Mathematical models with optimal control analysis are
an important tool in understanding the corruption transmis-
sion dynamics and in decision-making processes regarding
intervention programs for corruption control. In the current
research, there are only a few quantitative studies on corrup-
deterministic model for corruption in a population. They com-
puted the basic reproduction number (BRN), corruption-free
equilibrium point, and endemic equilibrium point. Numerical
simulations were carried out, and they revealed that corruption
can only be reduced to a manageable level but not totally
removed. In [2], the authors developed and analysed a math-
ematical model for corruption dynamics. They determined
the basic reproduction number and corruption-free and
endemic equilibrium points. The study by Lemecha [8]
proposed a mathematical model for corruption by consider-
ing awareness created by anticorruption and counseling in
jail. The existence of unique corruption-free and endemic
equilibrium points was investigated, and the basic reproduc-
tion number was computed. The study in [9] developed the
differential equation-based models representing either growth
or decay laws for corruption. In [10], a difference equation-
based model was framed to measure the level of corruption.
The author in [11] considered a game theoretic approach to
deal with corruption. A corruption prevention model was
formulated by Khan [12] and showed that the complete
prevention of corruption is possible if the ratio between rate
2 Journal of Applied Mathematics

of dismissal and rate of corruption is equal to one. In [13], the With regard to the above considerations, we have the
authors developed an SIR model for the corruption dynamics. compartmental flow diagram shown in Figure 1. From the
Furthermore, they extended the model to include optimal flow chart, the model will be governed by the following
control with a single optimal control strategy. They concluded system of differential equations:
that the level of corruption in society can be reduced if efforts
to control corruption are increased and put in place through dS
media/punishments. Nathan and Jackob developed an epide- = Π + ð1 − θÞεR − pβSC − ðk + μÞS,
dt
miological compartment model for corruption in Kenya,
dE
mainly considering those who take advantage of office holders = pβSC − ðδ + μÞE,
and political office holders [14]. dt
In this paper, an optimal control theory was used to study dC
= αδE − ðσ + μÞC, ð1Þ
the effectiveness of all possible combinations of two corruption dt
preventive measures, namely, (i) campaigning about corrup- dR
tion through media and advertisement and (ii) exposing the = σC + ð1 − αÞδE − ðε + μÞR,
dt
corrupted individuals to jail and giving punishment. For this,
we considered a model for the transmission of corruption dH
= κS + θεR − μH,
and incorporated two time-dependent controls. Firstly, the dt
model has been studied with no control and the stability prop-
erties have been investigated. Secondly, we consider the control with the initial condition
parameters to be time dependent and examine the impact of
different combinations of these measures in controlling Sð0Þ = S0 ≥ 0,
corruption. We use Pontryagin’s maximum principle to derive
Eð0Þ = E0 ≥ 0,
necessary conditions for the optimal control of corruption.
The paper is organized as follows. Section 2 is devoted for I ð0Þ = I 0 ≥ 0, ð2Þ
the model description and the underlying assumptions. In
Section 3, we carry out mathematical analysis including Rð0Þ = R0 ≥ 0,
bifurcation and sensitivity of the model. In Section 4, the H ð0Þ = H 0 ≥ 0:
optimal control problem is presented. Section 5 is devoted
for numerical simulations and discussions of the model.
The conclusion is presented in Section 6. 3. Model Analysis
3.1. Positivity of Solutions. Since the model (1) monitors
2. Model Formulation and Description human population, it is important that all its state variables
and associated parameters are positive for future time. This
The total population NðtÞ is divided into five compartments. will be established by the following theorem:
Those who are susceptible to corruption are susceptible
individuals SðtÞ, those who are exposed to a corrupted person Theorem 1. Let Ω = fðS, E, C, R, HÞ ∈ R5 : Sð0Þ > 0, Eð0Þ >
but do not perform it are exposed individuals EðtÞ, those who 0, Cð0Þ > 0, Rð0Þ > 0, Hð0Þ > 0g. Then, the solution set ðSðtÞ,
are performing corruption are corrupted individuals CðtÞ, EðtÞ, CðtÞ, RðtÞ, HðtÞÞ of system (1) is positive for all t ≥ 0.
those who stopped doing corruption are recovered individ-
uals RðtÞ, and those who know the badness of corruption Proof. We let τ = sup ft > 0 : S0 ðνÞ ≥ 0, E0 ðνÞ ≥ 0, C0 ðνÞ ≥ 0,
and do not perform it permanently are honest individuals R0 ðνÞ ≥ 0, H 0 ðνÞ ≥ 0 for all ν ∈ ½0, tg. Since S0 ðtÞ ≥ 0, E0 ðtÞ ≥
HðtÞ at time t ≥ 0. Assume that there is a positive recruitment 0, C 0 ðtÞ ≥ 0, R0 ðtÞ ≥ 0 and H 0 ðtÞ ≥ 0, then τ > 0. If τ < ∞,
Π into the susceptible class by birth or immigration. From then automatically S0 ðtÞ or E0 ðtÞ or C 0 ðtÞ or R0 ðtÞ or H 0 ðtÞ
this class, κ proportion will join the honest subpopulation is equal to zero at τ. Taking the first equation of the model (1)
which never involve in corruption practices irrespective of
the circumstances around them. We consider a positive dS
natural death rate μ for all humans at all time under the = Π + ð1 − θÞεR − pβSC − ðk + μÞS: ð3Þ
study. Susceptible individuals will have a contact rate with dt
the corrupted individuals at rate β with a corruption trans-
Then, using the variation of constant formula the solution
mission probability p per contact and moved to the exposed
of equation (3) at τ is given by
class. From these individuals, α progressed at a rate of δ to
the corrupted compartment and the remaining proportion  ðT 
will move to the recovered compartments. The corrupted SðτÞ = Sð0Þ exp − ðpβC + ðk + μÞÞðsÞds
individuals get knowledge about the effect of corruption in 0
ðt
jail and move to the recovered subpopulation at a rate of σ.
From these recovered individuals, θ moved at a rate of ε to + ðΠ + ð1 − θÞεRÞ ð4Þ
0
susceptible compartment and the other proportion joins the  ðt 
honesty compartment. All the descriptions of the parameters ⋅ exp − ðpβC + ðk + μÞÞðvÞdv ds > 0:
are listed in Table 1. s
Journal of Applied Mathematics 3

Table 1: Description of parameters of the corruption model (1).
Parameter Description
p Corruption transmission probability per contact
β Contact rate of corrupted individual to the susceptible
individual
δ Rate at which exposed individuals become corrupted
σ Rate at which corrupted individuals become recovered
ε Rate at which recovered individuals become honest
Λ Recruitment rate of susceptible humans
κ Proportion of individuals that joins the honest
population from susceptible population
μ Death rate of all humans
Proportion of individuals that joins the corrupted
α introduced into a susceptible population. We calculate
subpopulation from the exposed compartment
θ Proportion of individuals that joins the honest
subpopulation from the recovered compartment
is positively invariant, inside which the model is considered
to be epidemiologically meaningful and mathematically well
posed.
3.3. Corruption-Free Equilibrium Point (CFEP). The
corruption-free equilibrium point of model (1) is obtained
by equating all equations of the model to zero and letting
E = 0, C = 0, and R = 0. Then, we get
 
Π kΠ
E0 = , 0, 0, 0, : ð8Þ
k+μ μðk + μÞ
3.4. Basic Reproduction Number. The basic reproduction
number R0 measures the expected number of secondary
infections that result from one newly infected individual
R0 of the model using the next-generation matrix method
as described in [15]. The first step to get R0 is rewriting
the model equations starting with newly infective classes:

𝜋 𝛽 𝛼𝛿
S E C dE
= pβSC − ðδ + μÞE,
dt
𝜇 𝜇 𝜇 dC
= αδE − ðσ + μÞC,
𝜅 (1 – 𝛼)𝛿 𝜎 dt
(1 – 𝜃)𝜀 ð9Þ
dR
= σC + ð1 − αÞδE − ðε + μÞR,
dt
𝜃𝜀
H
dH
R = κS + θεR − μH:
dt
𝜇 𝜇

Figure 1: Compartmental diagram for the transmission dynamics Then, by the principle of next-generation matrix, the
of corruption. Jacobian, at CFEP, we obtained
0 Πβp 1
Moreover, since all the variables are positive in ½0, τ, hence 0 0 0
B k+μ C
B C
SðτÞ > 0. It can be shown in a similar way that EðτÞ > 0, CðτÞ B0 0C
> 0, RðτÞ > 0, and HðτÞ > 0, which is a contradiction. Hence, F=B
B
0 0 C,
C
τ = ∞. Therefore, all the solution sets are positive for t ≥ 0. B0 0C
@ 0 0 A
3.2. Invariant Region. We obtained the invariant region, in 0 0 0 0 ð10Þ
which the model solution is bounded. To do this, consider 0 1
the total human population NðtÞ, where NðtÞ = SðtÞ + EðtÞ + δ+μ 0 0 0
B C
CðtÞ + RðtÞ + HðtÞ. B −αδ σ+μ 0 0C
V =B
B
C:
C
B −ð1 − αÞδ −σ ε+μ 0C
dN dS dE dC dR dH
= + + + + : ð5Þ @ A
dt dt dt dt dT dt 0 0 −θε μ
Substituting all state equations from the model (1), we get
Therefore, the basic reproduction number is the spectral
dN radius of the next generation matrix FV −1 and is given by
= Π − μN: ð6Þ
dt
pβδαΠ
Integrating equation (6) both sides and taking t ⟶ ∞, R0 = : ð11Þ
ðδ + μÞðσ + μÞðk + μÞ
we get
  3.5. Local Stability of CFEP
Π
Ω = ðS, E, C, R, H Þ ∈ R5+ : Sðt Þ + Eðt Þ + C ðt Þ + Rðt Þ + H ðt Þ+≤
μ
Theorem 2. The CFEP is locally asymptotically stable if
ð7Þ R0 < 1 and unstable if R0 > 1.
4 Journal of Applied Mathematics

Proof. To prove this theorem, let us first find the Jacobian −Παβδp + ðδ + μÞðκ + μÞðσ + μÞ
matrix of system (1): ψ2 =
κ+μ ð18Þ
0 1 = ðδ + μÞðσ + μÞð1 − R0 Þ > 0:
−βpC − ðκ + μÞ 0 −pβS ð1 − θÞε 0
B C
B βpC −δ − μ pβS 0 0 C
B C Hence, the CFEP is locally asymptotically stable if R0 < 1.
B C
J =B
B 0 αδ −μ − σ 0 0 C
C:
This result implies that corruption can be eliminated if
B C the initial size of the corrupted individuals is in the basin of
B 0 ð1 − αÞδ σ − ðε + μ Þ 0 C
@ A attraction of the CFEP. On the other hand, to ensure that
corruption elimination is independent of the initial popula-
κ 0 0 θε −μ
tion size of the corrupted population, it is necessary to show
ð12Þ that the CFEP is globally stable.

Evaluating equation (12) at the corruption-free 3.6. Global Stability of CFEP

equilibriumE0 = ðΠ/ðk + μÞ, 0, 0, 0, Π/μðk + μÞÞ, we got
Theorem 3. The equilibrium point E0 of the model (1) is
0 1 globally asymptotically stable if R0 < 1.
pβΠ
B −ðκ + μÞ 0 −
κ+μ
ð1 − θÞε 0 C
B C Proof. Consider the following Lyapunov function
B C
B pβΠ C
B
B 0 −δ − μ 0 0 C
C
κ+μ C: V = χ1 E + χ2 C: ð19Þ
J =B
B C
B
B 0 αδ −μ + σ 0 0 C
C
B C Differentiating equation (19) with respect to t gives
B
@ 0 ð1 − α Þδ σ −ðε + μÞ 0 CA
κ 0 0 θε −μ dV dE dC
= χ1 + χ2 : ð20Þ
ð13Þ dt dt dt

From the Jacobian matrix, a characteristic polynomial Substituting dE/dt and dC/dt from the model (1), we get
was obtained as
dV
 = χ1 ½ðpβδC Þ − ðδ + μÞE + χ2 ½αδE − ðσ + μÞC 
ð−λ − ðκ + μÞÞð−λ − μÞð−λ − ðε + μÞÞ λ2 + ψ1 λ + ψ2 = 0, dt ð21Þ
= χ1 pβδC − χ2 ðσ + μÞC − χ1 ðδ + μÞE + χ2 αδE:
ð14Þ

where Here, take χ1 = ðαδ/ðδ + μÞÞχ2 , then we have

ψ1 = σ + 2μ + δ, dV αδ
= χ βδC − χ2 ðσ + μÞC
ð15Þ dt δ + μ 2
−Παβδp + ðδ + μÞðκ + μÞðσ + μÞ   ð22Þ
ψ2 = : pαβδΠ
κ+μ ≤ − ðσ + μÞ χ2 C :
ðδ + μÞðk + μÞ
From equation (14), we have
Taking χ2 = 1 and substituting R0 , we get
−λ − ðκ + μÞ ⇒ λ1 = −ðκ + μÞ < 0,
dV
−λ − μ ⇒ λ2 = −μ < 0, ð16Þ ≤ ðσ + μÞðR0 − 1ÞC, ð23Þ
dt
−λ − ðε + μÞ ⇒ λ3 = −ðε + μÞ < 0:
for C ≤ C0 = π/ðk + μÞ and dV/dt ≤ 0 for R0 < 1 and dV/
From the last expression of equation (14), we have dt = 0 if and only if C = 0. This implies that the only trajec-
tory of the system (1) on which dV/dt ≤ 0 is E0 . Therefore,
λ2 + ψ1 λ + ψ2 = 0: ð17Þ by LaSalle’s invariance principle, E0 is globally asymptoti-
cally stable in Ω. This supports the forward bifurcation of
Section 3.9.
The Routh-Hurwitz criteria was applied, and equation
(17) has strictly negative real root iff ψ1 > 0 and ψ2 > 0. 3.7. The Endemic Equilibrium Point. The endemic equilib-
Clearly, we see that ψ1 > 0 because it is the sum of positive rium point denoted by E∗ = ðS∗ , E∗ , C ∗ , R∗ , H ∗ Þ:E∗ is the
parameters and also steady state solution where corruption persists in the
Journal of Applied Mathematics 5

population. It can be obtained by equating each equation ðε + μÞ½R0 − 1

in (1) to zero: ψ3 = −ðδ + μÞðσ + μÞðε + κ + 3μÞ + fδð1
ϕ
− θ + σ + 3μÞ + ε + 3μð2μ + σÞ − αδðθ − 1Þg
dS dE dC dR dH
= = = = = 0: ð24Þ + ðσ + 3μÞ½δκ + εκ + δε + ð2μ + σÞ½3εμ − 3κμ − 6δμ
dt dt dt dt dt
+ δεκ + 2μ2 ð5μ + σÞ,
Then, we obtain
ψ4 = 2βpC½αδθðσ + μÞ + αδðμ + σÞðθ + 1Þ + ðσ + μÞδðε + 1Þ
+ βpC ½δμð3μ + 2σ − 2θÞ + δσθ + εμð3μ + 2σÞ
ðδ + μÞðσ + μÞ
S∗ = , + μ2 ð4μ + 3σÞ + 2ðμ + σÞδεκ + ð3μ + 2σÞ½δεμ + δκp
αβδp
+ εκp − μ2 ð4μ + 3σÞ½ε + κ + δ + μ3 ð5μ + 4σÞ
ðσ + μÞðε + μÞ½1 − R0 

E∗ = , − ðδ + μÞðσ + μÞ εκ + 2εμ + 2κμ + 3μ2 ,
αδpβϕðδ + μÞðσ + μÞðκ + μÞ "
ðε + μÞ½1 − R0  −μðε + μÞ½R0 − 1
C∗ = , ψ5 = ðμ + 2σÞαδðθ − 1Þ
pβϕðδ + μÞðσ + μÞðκ + μÞ ϕ

½μðα − 1Þ − σ½R0 − 1 − ðσ + μÞ½βδðε + μ − θÞ + βμðε + μ + δÞ

R∗ = ,  
αpβϕðδ + μÞðσ + μÞðκ + μÞ ε
− ðκ + μÞðδ + μÞðσ + μÞμ βμ + βκμ
β
κðδ + μÞðσ + μÞ εθ½μðα − 1Þ − σ½R0 − 1 
H∗ = + , + ðσ + μÞμ εκðβ + μÞ + μδðκ + εÞ
αδpβμ αpβϕμðδ + μÞðσ + μÞðκ + μÞ #
ð25Þ
+ μ ðδ + ε + κ + μÞ ,
2

where ð29Þ

ϕ = αδμðθ − 1Þ − ðε + μÞðδ + μÞðσ + μÞ − δð1 + θÞðσ + μÞ: Using the Routh-Hurwitz criterion, all roots of char-
acteristic polynomial have negative real parts if and only
ð26Þ if ψ1 > 0, ψ2 > 0, ψ3 > 0, ψ4 > 0, ψ5 > 0, ψ1 ψ2 ψ3 > ψ23 + ψ21 ψ4
3.8. Local Stability of Endemic Equilibrium and ðψ1 ψ4 − ψ5 Þðψ1 ψ2 ψ3 − ψ23 + ψ21 ψ4 Þ > ψ5 ðψ1 ψ2 − ψ3 Þ2 +
ψ1 ψ25 for R0 > 1. Hence, the endemic equilibrium E∗ is
Theorem 4. The endemic equilibrium E∗ of system (1) is locally asymptotically stable.
locally asymptotically stable in Ω if R0 > 1, otherwise unstable.
3.9. Bifurcation Analysis. In order to determine the direction
of the bifurcation and to prove the stability of the endemic
Proof. Let us first obtain the Jacobian matrix of system (1):
equilibrium point, we make use of the bifurcation theory,
0 1 which is based on the center manifold theory [16] (as
−βpC − ðκ + μÞ 0 −pβS ð 1 − θÞ ε 0 described by Theorem 6 of Castillo-Chavez and Song [16]).
B C
B βpC −ðδ + μÞ pβS 0 0 C
B C
B C Theorem 5. The model in the system (1) exhibits forward
J =B
B 0 αδ −ðμ + σÞ 0 0 C
C: bifurcation at R0 = 1 and a unique endemic equilibrium exists.
B C
B 0 ð 1 − αÞ δ σ −ðε + μÞ 0 C
@ A
Proof. To conduct this analysis, we assume that S = z1 ,
κ 0 0 θε −μ E = z 2 , C = z 3 , R = z 4 , H = z 5 and from R0 = 1, we have
ð27Þ
ðδ + μÞðσ + μÞðk + μÞ
β = β∗ = : ð30Þ
The characteristic polynomial of equation (27) is given by pδαΠ
The new the model in equation (1) is given by
λ5 + ψ1 λ4 + ψ2 λ3 + ψ3 λ2 + ψ2 λ + ψ4 = 0, ð28Þ
dz 1
= Π + ð1 − θÞεz 4 − pβz 1 z 3 − ðk + μÞz 1 ,
where dt
dz 2
= pβz 1 z 3 − ðδ + μÞz 2 ,
ðσ + μÞðε + μÞ½R0 − 1 dt
ψ1 = + δ+ ∈ +κ + 5μ + σ, dz 3
αδϕ = αδz 2 − ð1 − μÞz 3 , ð31Þ
dt
  dz 4
ðε + μÞ½R0 − 1 = σz 3 + ð1 − αÞδz 2 − ðε + μÞz 4 ,
ψ2 = −ðδ + μÞðσ + μÞ + ðδ + ε + σ + 4μÞ +δ dt
ϕ
dz 5
+ ð4μ+σÞð∈+κÞ + εκ + μð10μ + 4σÞ, = κz 1 + θεz 4 − μz 5 :
dt
6 Journal of Applied Mathematics

The Jacobian of the model system (31) around the Table 2: Sensitivity index table.
corruption-free equilibrium point E0 = ðz 1 = Π/ðκ + μÞ, 0,
0, 0, z 5 = κΠ/μðκ + μÞÞ evaluated at R0 = 1 is Parameter symbol Sensitivity indices

1 β +ve
0
pβΠ δ
B −ðκ + μÞ − ð1 − θÞε
0 0 C +ve
B κ+μ C
B C α +ve
B pβΠ C
B
B 0 −δ − μ 0 0 C
C
κ -ve
J =B κ+μ C: ð32Þ
B C μ -ve
B
B 0 αδ −μ + σ 0 0 C
C
B C σ -ve
B
@ 0 ð1 − αÞδ σ −ðε + μÞ 0 C
A
κ 0 0 θε −μ 2v2 w22 pβαδ
a = 2v2 w1 w3 = ðL − W Þ < 0,
ðσ + μÞ2
The right eigenvector, w = ðw1 , w2 , w3 , w4 , w5 ÞT , asso- ð36Þ
ciated with this simple zero eigenvalue can be obtained δαΠ
b = v2 w3 pz∗1 = v2 > 0,
from Jw = 0. ðk + μÞðσ + μÞ
 
ð1 − θÞε½ð1 − αÞδðα + μÞ + σδα δ + μ where
w1 = − w,
ðσ + μÞðε + μÞðκ + μÞ κ+μ 2
w2 = w2 > 0, εδ½ð1 − θÞðσ + μÞ + θαð2σ + μÞ
L= ,
ðε + μÞðκ + μÞ
αδ ð37Þ
w3 = w,
μ+σ 2 εαδμ + ðε + μÞðδ + μÞ2
W= :
ð1 − αÞεðσ + μÞ + σδα ðε + μÞðκ + μÞ
w4 = w2 ,
ðμ + σÞðε + μÞ
  Since the coefficient b is positive and the sign of the
κ ð1 − θÞε½ð1 − αÞδðσ + μÞ + σδα δ + μ coefficient a is negative, the model undergoes forward
w5 = − w
μ ðσ + μÞðε + μÞðκ + μÞ κ+μ 2 bifurcation, which implies that R0 < 1 is a necessary
θε½ð1 − αÞδðσ + μÞ + σδα and sufficient condition for corruption elimination.
+ w2 :
μðμ + σÞðε + μÞ 3.10. Sensitivity Analysis. In order to see the relative effect of
ð33Þ each parameter to the corruption transmission, we performed
sensitivity analysis. To go through, we followed the approach
The left eigenvector, v = ðv1 , v2 , v3 , v4 , v5 Þ, associated defined by [17] as done in [18, 19] which used the definition
with this simple zero eigenvalue can be obtained from of normalized forward sensitivity index defined as a variable,
vJ = 0 and given by R0 , that depends differentiably on a parameter,ℓ, defined as

v1 = v4 = v5 = 0, R ∂R0 ℓ
Λℓ 0 = × , ð38Þ
∂ℓ R0
v2 = v2 > 0, ð34Þ
δ+μ for ℓ represents all the basic parameters and R0 = αpβδΠ
v3 = v : /ðδ + μÞðσ + μÞðκ + μÞ.
αδ 2

Since the first, fourth, and fifth components of v are ∂R0 β

ΛR
κ =
0 × = 1 ≥ 0,
zero, we do not need the derivatives of f 1 , f 4 , and f 5 . ∂β R0
From the derivatives of f 2 and f 3 , the only ones that
∂R0 β κ
are nonzero are the following: ΛR
κ =
0 × =− ≤ 0, ð39Þ
∂κ R0 κ+μ
∂2 f 2 ∂2 f 2 ∂R0 β σ
= = pβ, ΛR
σ =
0 × =− ≤ 0:
∂z 2 ∂z 3 ∂z 3 ∂z2 ∂σ R0 σ+μ
ð35Þ
∂2 f 2
= pz∗1 , And we do this in a similar fashion for the remaining
∂z 2 ∂β parameters.
and all the other partial derivatives are zero. The direc- 3.11. Interpretation of Sensitivity Indices. The sensitivity
tion of the bifurcation at R0 = 1 is determined by the indices of parameters are found in Table 2. Parameters that
signs of the bifurcation coefficients a and b, obtained have positive indices ðβ, α, and δÞ have great impact on
from the above partial derivatives, given, respectively, by expanding the corruption in the community when their
Journal of Applied Mathematics 7

values increase. Parameters with negative indices ðκ, σ, and μÞ
minimize the burden of corruption in the community as their
values increase.
Therefore, the model sensitivity analysis demonstrated
that policy makers are supposed to decrease positive index
parameters and increase negative index parameters to com-
bat corruption in a population. From this result, to identify
the optimal strategy to fight against corruption, an optimal
control model analysis is considered in the next section.
4. Extension of the Model into Optimal Control
The optimal level that would be needed to control corruption
is minimizing the number of exposed and corrupted popula-
tion. An optimal control model was developed by including
the following two controls:
4.1. The Hamiltonian and Optimality System. The optimal
control must satisfy the necessary conditions that are formu-
lated by Pontryagin’s maximum principle ([21]). This princi-
ple converts the system of equations (40) and (41) into a
problem of minimizing point-wise a Hamiltonian (M), with
respect to u1 ðtÞ and u2 ðtÞ as
1
M = a1 E + a2 C + w1 u21 + w2 u22 + w3 u23
2
+ λ1 ½Π + ð1 − θÞεR − ð1 − u2 ÞpβSC − ðk + μ + u1 ÞS
+ λ2 ½ð1j−u2 ÞpβSC − ðδ + μ + u1 ÞE + λ3 ½ð1 − u1 ÞαδE
− ðσ + μ + u2 ÞC + λ4 ½ðσ + u2 ÞC + ð1 − αÞðδ + u2 ÞE
− ðε + μÞR + λ5 ½ðk + u1 ÞS + θεR − μH ,
ð43Þ

(i) u1 : campaigning about corruption through media where λi , i = 1, ⋯, 5 are the adjoint variable functions to be
and advertisement determined. Because of the convexity of the integrand of J
with respect to u1 and u2 , a priori boundaries of the state
(ii) u2 : exposing the corrupted individuals to jail and solutions and the Lipschitz property of the state system relat-
giving punishment ing to the state variables, the existence of the optimal control
was proved using Corollary 4.1 in the work of Fleming and
Adding the two controls on the model (1), the optimal
Rishel [22].
control model is given by
Theorem 6. Suppose we have optimal controls u∗1 , u∗2 and
dS S, E, C, R, H solutions of the respective state system that
= Π + ð1 − θÞεR − ð1 − u2 ÞpβSC − ðk + μ + u1 ÞS,
dt minimizes J over U, there exist adjoint variables, λ1 , ⋯,
dE λ5 such that
= ð1 − u2 ÞpβSC − ðδ + μ + u1 ÞE,
dt
dλ1
dC = λ1 ðð1 − u2 ÞpβC + κ + μ + u1 Þ
= ð1 − u1 ÞαδE − ðσ + μ + u2 ÞC, dt
dt − λ2 ð1 − u2 ÞpβC − λ5 ðκ + u1 Þ,
dR
= ðσ + u2 ÞC + ð1 − αÞðδ + u2 ÞE − ðε + μÞR,
dt dλ2
= −a1 + λ2 ðδ + μ + u1 Þ − λ3 αδð1 − u1 Þ
dH dt
= ðk − u1 ÞS + θεR − μH:
dt − λ4 ð1 − αÞðδ + u1 Þ,
ð40Þ
dλ3
= −a2 + λ1 ð1 − u2 ÞpβS − λ2 ð1 − u2 ÞpβS
The control variables u1 and u2 minimize the optimal con- dt
trol model (40) subject to the objective functional defined as + λ3 ðσ + μ + u2 Þ − λ4 ðσ + u2 Þ,
ðt f  
1
 dλ4
J= 2 2
a1 E + a2 I + w1 u1 + w2 u2 dt, ð41Þ = λ1 ∈ ð1 − θÞ+λ4 ð∈+μÞ − λ5 θ ∈ ,
0 2 dt
dλ5
= μλ5 : ð44Þ
where t f is the final time, a1 and a2 are weight constants of the dt
exposed and corrupted population, respectively, while w1 and
w2 are weight coefficients for each individual control measure. With transversality conditions, λi ðt f Þ = 0, i = 1, ⋯, 5.
We choose a nonlinear cost on the controls based on the Furthermore, we obtained the control set ðu∗1 , u∗2 Þ charac-
assumption that the cost takes nonlinear form [20]. Optimal terized by
control function ðu∗1 , u∗2 Þ needs to be found such that   
ðλ − λ5 ÞS + ½λ2 + αδλ3 − ð1 − αÞλ4 E
u∗1 = max 0, min 1, 1 ,
w1
J ðu∗1 , u∗2 Þ = min f J ðu1 , u2 Þjðu1 , u2 Þ ∈ U g, ð42Þ   
pβCSðλ2 − λ1 Þ + ðλ4 − λ3 ÞC
u∗2 = max 0, min 1, :
w2
where U = fðu1 , u2 Þjui ðtÞ is lebesgue measurable on ½0, t f ,
0 ≤ ui ðtÞ ≤ 1, i = 1, 2g is the closed set. ð45Þ
8 Journal of Applied Mathematics

Proof. The adjoint equation and transversality conditions are Table 3: Values of parameter used in the numerical simulation.
standard results obtained from Pontryagin’s maximum prin-
ciple [21]. Differentiate the Hamiltonian with respect to Parameter Value Source
states S, E, C, R, and H, respectively, then we got the adjoint Π 85 Assumed
equations as p 0.036 [7, 8]
β 0.0234 [7, 8]
dλ1 ∂M
=− = λ1 ðð1 − u2 ÞpβC + κ + μ + u1 Þ δ 0.2 Assumed
dt ∂S
− λ2 ð1 − u2 ÞpβC − λ5 ðκ + u1 Þ, σ 0.007 Assumed
ε 0.35 Assumed
dλ2 ∂M κ 0.03 [8]
− = −a1 + λ2 ðδ + μ + u1 Þ − λ3 αδð1 − u1 Þ
dt ∂E μ 0.0160 [8]
− λ4 ð1 − αÞðδ + u1 Þ, α 0.3 [8]
θ 0.1 [8]
dλ3 ∂M
− = −a2 + λ1 ð1 − u2 ÞpβS − λ2 ð1 − u2 ÞpβS
dt ∂C
+ λ3 ðσ + μ + u2 Þ − λ4 ðσ + u2 Þ, system by incorporating the characterized control set under
initial and transversal condition:
dλ4 ∂M 8
dS Y
− = −λ1 ∈ ð1 − θÞ+λ4 ð∈+μÞ − λ5 θ ∈ , >
>
> = +ð1 − θÞεR − ð1 − u∗2 ÞpβSC − ðk + μ + u∗1 ÞS,
dt ∂R > dt
>
>
>
>
> dE
>
dλ5 ∂M >
>
> = ð1 − u∗2 ÞpβSC − ðδ + μ + u∗1 ÞE,
− = μλ5 : > dt
>
>
dt ∂H >
> dC
>
>
> = 1 − u∗1 ÞαδE − ðα + μ + u∗2 ÞC,
ð46Þ > dt ð
>
>
>
>
>
>
>
> dR
= ðσ + u∗2 ÞC + ð1 − αÞðδ + u∗2 ÞE − ðε + μÞR,
>
>
With transversality conditions, λiðtf Þ = 0, i = 1, ⋯, 5. >
>
>
dt
>
>
Differentiate the Hamiltonian equation with respect to u1 >
> dH = ðk + u∗ ÞS + θεR − μH,
>
>
and u2 and obtain the characterization of optimal controls >
> dt 1
>
>
>
u∗1 , u∗2 from ∂M/∂u1 = 0, ∂M/∂u2 = 0. Thus, we got the char- >
>
>
>
dλ1
= λ1 ðð1 − u∗2 ÞpβC + k + μ + u∗1 Þ − λ2 ð1 − u∗2 ÞpβC − λ5 ðk + u∗1 Þ,
>
> dt
acteristic equation in standard control arguments form >
>
>
>
involving the bounds on the controls as follows: >
> dλ2 = −a + λ ðδ + μ + u∗ Þ − λ αδð1 − u∗ Þ − λ ð1 − αÞðδ + u∗ Þ,
>
>
>
> dt
1 2 1 3 1 4 1
>
8 >
>
>
> dλ
>
> ψ , if 0 < ψ1 < 1, >
>
>
3
= −a2 + λ1 ð1 − u∗2 ÞpβS − λ2 ð1 − u∗2 ÞpβS + λ3 ðσ + μ + u∗2 Þ − λ4 ðσ + u∗2 Þ,
< 1 >
>
<
dt
∗
u1 = 0, if ψ1 ≤ 0, dλ4
>
> >
> = −λ1 ∈ ð1 − θÞ+λ4 ð∈+μÞ − λ5 θε,
: > dt
>
>
1, if ψ1 ≥ 1, >
>
> dλ5
8 ð47Þ >
> = μλ5 ,
>
> dt
> ψ2 , if 0 < ψ2 < 1, >
>
> >
>   
< > ∗
>
>
ðλ1 − λ5 ÞS + ½λ2 + αδλ3 − ð1 − αÞλ4 E
> u = max 0, min 1, ,
u∗2 = 0, if ψ2 ≤ 0, >
> 1
>
>
w1
>
> >
>   
: >
>
> pβCSðλ2 − λ1 Þ + ðλ4 − λ3 ÞC
1, if ψ2 ≥ 1: >
>
>
u∗2 = max 0, min 1, ,
>
> w2
>
>

>
> λi t f = 0,
In compacted notation: >
>
>
>
>
> i = 1, ⋯, 5,
>
>
>
>
u∗1 = max f0, min ð1, ψ1Þg, >
>
>
> Sð0Þ = S0 ,
ð48Þ >
>
>
>
u∗2 = max f0, min ð1, ψ2Þg, >
>
>
>
Eð0Þ = E0 ,
>
>
> C ð0 Þ = C 0 ,
>
>
>
where >
>
>
> Rð0Þ = R0 ,
>
>
:
H ð0Þ = H 0 :
ðλ1 − λ5 ÞS + ½λ2 + αδλ3 − ð1 − αÞλ4 E
ψ1 = , ð50Þ
w1
ð49Þ
pβCSðλ2 − λ1 Þ + ðλ4 − λ3 ÞC
ψ2 = : 5. Numerical Simulations and Discussions
w2
In this section, numerical simulations are performed on the
The optimality system is formed from the optimal model (1). Maple 18 software is used to check the effect of some
control system (the state system) and the adjoint variable parameters in the spread and control of corruption. The
Journal of Applied Mathematics 9

7000

6000

Corrupted population
5000

4000

3000

2000

1000

0 5 10 15
Time (years)

𝛼 = 0.6 𝛼 = 0.9
𝛼 = 0.4 𝛼 = 0.8
𝛼 = 0.3
(a)

20000
Corrupted population

15000

10000

5000

0 5 10 15
Time (years)

𝛼 = 0.6 𝛼 = 0.7
𝛼 = 0.8 𝛼 = 0.4
𝛼 = 0.2
(b)

Figure 2: Effect of α and δ on the corrupted population.

parameter values in Table 3 and the initial conditions Sð0Þ =
10000, Eð0Þ = 0, Cð0Þ = 100, Rð0Þ = 0, Hð0Þ = 100 are used.
5.1. Effect of α and δ on the Corrupted Population. As we see
in Figure 2(a), we experimented the effect of changing α on
the number of corrupted population by keeping the other
parameters constant. It can be seen from Figure 2(a) that,
as the α value increases, the number of corrupted population
increases. Similarly, in Figure 2(b), we investigated the effect
of changing δ on the number of corrupted population by
10 Journal of Applied Mathematics

4000

Corrupted population
3000

2000

1000

0 5 10 15
Time (years)

𝜅 = 0.09 𝜅 = 0.2
𝜅 = 0.4 𝜅 = 0.06
𝜅 = 0.03
(a)

4000
Corrupted population

3000

2000

1000

0 5 10 15
Time (years)

𝜎 = 0.06 𝜎 = 0.09
𝜎 = 0.1 𝜎 = 0.02
𝜎 = 0.007
(b)

Figure 3: Effect of κ and σ on the corrupted population.

keeping the other parameters constant. It is evident from
Figure 2(b) that, when the value of δ increases, the number
of corrupted population also increases. Therefore, public
policy makers must concentrate on minimizing the contact
rate of corrupted individuals to susceptible population and
reduce the transition of exposed individuals to the corrupted
population.
5.2. Effect of κ and σ on the Corrupted Population. In
Figure 3(a), we can see that κ plays an important role in
Journal of Applied Mathematics 11

120 103

100 102

Corrupted population
Exposed population

101
80
100

60
99

40 98

97
20

96
0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (years) Time (years)
u1 = u2 = 0
u1 ≠ 0,u2 = 0
(a) (b)

Figure 4: Simulations of the corruption model showing the effect of the optimal strategies u1 ≠ 0.

700 150

600
140

500
Corrupted population
Exposed population

130
400

300 120

200
110
100

0 100
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (years) Time (years)

(a) (b)

Figure 5: Simulations of the corruption model showing the effect of the optimal strategies u2 ≠ 0:

reducing the number of corrupted population. When the rate
at which individuals join the honest subpopulation from sus-
ceptible class increases from 0.03 to 0.4, the number of cor-
rupted population decreases. From Figure 3(b), the rate at
which corrupted individuals join the recovered population
through education and/or jail sentence increases from 0.007
to 0.1; the number of corrupted population decreases. There-
fore, public policy makers must concentrate on maximizing
the number of populations that join the honest subpopula-
tion and corrupted population to recover from doing it.
To simulate the optimality system in (50), iterative tech-
nique is applied which gives the optimal solution. Forward
fourth-order Runge-Kutta method is used to solve the state
system, and backward fourth-order Runge-Kutta method is
used to solve the adjoint system by considering the initial
condition of the state system and the final condition of the
12
500
400
Exposed population
300
200
100
0
0 0.5 1 1.5
Time (years)
(a)
Figure 6: Simulations of the corruption model showing the effect of the optimal strategies u1 ≠ 0 and u2 ≠ 0.
adjoint systems. The controls continue to be updated by
combining from the previous result of the controls with the
characterization. The solution of the state and adjoint system
is repeated by the updated controls. This condition continues
repeatedly until consecutive iteration is close enough each
other [23]. We proposed the following three strategies for
numerical simulation of the corruption model:
(i) Strategy A: campaigning through media and adver-
tisement about corruption ðu1 ≠ 0 and u2 ≠ 0Þ
(ii) Strategy B: exposing the corrupted individuals to jail
and giving punishment ðu1 ≠ 0 and u2 ≠ 0Þ
(iii) Strategy C: combination of both strategies A and B
ðu1 ≠ 0 and u2 ≠ 0Þ
5.3. Strategy A: Campaigning through Media and
Advertisement about Corruption. In this strategy, controls
u1 were used to optimize the objective function J while we
set control u2 to zero. In Figure 4, we see that exposed and
corrupted population significantly reduced when there is
control compared to situation with no control. From this,
we conclude that applying these strategy can reduce corrup-
tion along with time from the community but not possible to
avoid it in a specified period of time. Hence, it is not an
effective strategy to combat corruption from the population.
5.4. Strategy B: Exposing the Corrupted Individuals to Jail and
Giving Punishment. Under this strategy, we optimize the
objective function J using the control u2 while u1 is set to
zero. It is evident from Figure 5 that the number of exposed
individuals cannot reduce using the strategy and the
corrupted population seems eliminated in the first year but
grows in number due to the increase in number of exposed
Journal of Applied Mathematics
130
120
Corrupted population
110
100
2 0 0.5 1 1.5 2
Time (years)
(b)
individuals. Therefore, the strategy is not effective in fighting
and reducing corruption in the population even though it
reduces the number of corrupted individuals when compared
with control from without control.
5.5. Strategy C: Combination of Both Strategies A and B.
Under this strategy, both the control measures ðu1 and u2 Þ
are used to optimize the objective functional J. The results
of the applications of the strategy are shown in Figure 6. It
can be seen from Figure 6 that the exposed and corrupted
populations decrease as compared to no control. Therefore,
the intervention strategy is effective in bringing down the
exposed and corrupted population in the specified period of
time.
6. Conclusions
In this paper, a mathematical model for the transmission
dynamics of corruption in a population was formulated.
The basic reproduction number R0 was computed, and the
stability of equilibrium points was investigated. Through
Lyapunov’s theory, the corruption-free equilibrium point is
globally asymptotically stable whenever R0 < 1 was proven.
When R0 = 1, bifurcation analysis of the model was proven
and it exhibits forward bifurcation. The local stability of the
unique endemic equilibrium whenever R0 > 1 was demon-
strated. Using the definition of normalized forward sensitiv-
ity, the sensitivity parameters were determined. It can be
inferred that positive indices should be decreased while
increasing negative indices to minimize/control corruption
in a population. Then, the optimal control model was formu-
lated by adding two time-dependent controls (campaigning
about corruption through media and advertisement and
exposing the corrupted individuals to jail and giving
Journal of Applied Mathematics
punishment). Optimal control theory was used to establish
conditions under which the spread of corruption can be
stopped and to examine the impact of a possible combination
of these two controls on the transmission of corruption. The
characterization of the optimal control was obtained by the
application of the Pontryagin’s maximum principle. It is
evident from the results of the numerical simulations that
the integrated control strategy should be taken to fight
against corruption. The cost-effectiveness analysis of the
integrated strategy would be considered in a future study.
Data Availability
The data supporting this deterministic model are from
previous published articles and they have been duly cited in
this paper.
Conflicts of Interest
The author declares that there is no conflict of interest
regarding the publication of this paper.
Acknowledgments
I would like to thank University of Gondar for the support to
do this research.
References
[1] S. Bahoo, I. Alon, and A. Paltrinieri, “Corruption in interna-
tional business: a review and research agenda,” International
Business Review, vol. 29, no. article 101660, 2019.
[2] A. Cuervo-Cazurra, “Corruption in international business,”
Journal of World Business, vol. 51, no. 1, pp. 35–49, 2016.
[3] H. N. Pontell and G. Geis, International Handbook of White-
Collar and Corporate Crime, Springer, New York, USA, 2007.
[4] F. Y. Egudam, F. Oguntolu, and T. Ashezua, “Understanding
the dynamics of corruption using mathematical modeling
approach,” International Journal of Innovative Science, Engi-
neering & Technology, vol. 4, no. 4, pp. 2348–7968, 2017.
[5] F. Heimann and N. Z. Boswell, “The oecd convention: mile-
stone on the road to reform,” in New Perspectives on Combat-
ing Corruption, pp. 65–74, Transparency International,
Washington, 1998.
[6] P. Milner-Smyth, “Consequences of corruption,” HR Future,
vol. 2017, no. Jul 2017, pp. 30-31, 2017.
[7] S. Abdulrahman, “Stability analysis of the transmission
dynamics and control of corruption,” Pacific Journal of Science
and Technology, vol. 15, no. 1, pp. 99–113, 2014.
[8] L. Lemecha, “Modelling corruption dynamics and its analysis,”
Ethiopian Journal of Sciences and Sustainable Development,
vol. 5, no. 2, pp. 13–27, 2018.
[9] S. R. Waykar, “Mathematical modelling: a comparatively
mathematical study model base between corruption and devel-
opment,” IOSR Journal of Mathematics, vol. 6, no. 2, pp. 54–
62, 2013.
[10] S. R. Wayker, “Mathematical modelling: a study of corruption
in the society of India,” International Journal of Scientific and
Engineering Research, vol. 4, no. 7, pp. 2303–2318, 2013.
13
[11] P. V. Nikolaev, “Corruption suppression models: the role of
inspectors’ moral level,” Computational Mathematics and
Modeling, vol. 25, no. 1, pp. 87–102, 2014.
[12] M. A. U. Khan, “The corruption prevention model,” Journal of
Discrete Mathematical Sciences and Cryptography, vol. 3, no. 1-
3, pp. 173–178, 2000.
[13] S. Athithan, M. Ghosh, and X.-Z. Li, “Mathematical modeling
and optimal control of corruption dynamics,” Asian-European
Journal of Mathematics, vol. 11, no. 6, article 1850090, 2018.
[14] O. M. Nathan and K. O. Jackob, “Stability analysis in a math-
ematical model of corruption in Kenya,” Asian Research Jour-
nal of Mathematics, vol. 2019, pp. 1–15, 2019.
[15] P. Van den Driessche and J. Watmough, “Reproduction num-
bers and sub-threshold endemic equilibria for compartmental
models of disease transmission,” Mathematical Biosciences,
vol. 180, no. 1-2, pp. 29–48, 2002.
[16] C. Castillo-Chavez and B. Song, “Dynamical models of tuber-
culosis and their applications,” Mathematical Biosciences and
Engineering, vol. 1, no. 2, pp. 361–404, 2004.
[17] S. M. Blower and H. Dowlatabadi, “Sensitivity and uncertainty
analysis of complex models of disease transmission: an hiv
model, as an example,” International Statistical Review/Revue
Internationale de Statistique, vol. 62, pp. 229–243, 1994.
[18] H. T. Alemneh, O. D. Makinde, and D. M. Theuri, “Ecoepide-
miological model and analysis of MSV disease transmission
dynamics in maize plant,” International Journal of Mathemat-
ics and Mathematical Sciences, vol. 2019, pp. 1–14, 2019.
[19] H. T. Alemneh, O. D. Makinde, and D. M. Theuri, “Mathemat-
ical modelling of msv pathogen interaction with pest invasion
on maize plant,” Global Journal of Pure and Applied Mathe-
matics, vol. 15, no. 1, pp. 55–79, 2019.
[20] H. T. Alemneh, O. D. Makinde, and D. M. Theuri, “Optimal
control model and cost-effectiveness analysis of Maize streak
virus pathogen interaction with pest invasion in Maize plant,”
Egyptian Journal of Basic and Applied Sciences, vol. 7, no. 1,
pp. 180–193, 2020.
[21] L. S. Pontryagin, V. Boltyanskii, R. Gamkrelidze, and
E. Mischchenko, The Mathematical Theory of Optimal Pro-
cesses, Interscience, New York, The Mathematical Theory of
Optimal Processes, New York, 1962.
[22] W. H. Fleming and R. W. Rishel, “Deterministic and stochastic
optimal control,” Applications of Mathematics, vol. 1, 1976.
[23] S. Lenhart and J. T. Workman, Optimal Control Applied to Bio-
logical Models, Chapman and Hall/CRC, London, UK, 2007.